Recommendation system based on group profiles of personal taste

Abstract

This document describes a method and system for recommending items, such as beverages, that members of a group are likely to find appealing. When group members are identified, the system may identify one or more preference models for each member. Each preference model represents a pattern of dependency between characteristics of items that the member has rated and the member's ratings for those items. The system may develop a group preference profile by merging the patterns of dependency for each of the members into a group preference model. Then, when it receives a request for a recommendation for an item, the system uses the group preference profile to select, from a database, a candidate item having characteristics which are likely to appeal to many or all members of the group.

The invention claimed is: 
     
       1. A group recommendation system, comprising:
 a non-transitory computer-readable medium holding a database comprising characteristics for a plurality of candidate items; 
 a processor; and 
 a non-transitory computer-readable medium holding programming instructions that, when executed, are configured instruct the processor to:
 identify a preference profile for each user in a group of users, wherein the preference profile for each user comprises data that represents a pattern of dependency between the user's ratings for a plurality of items that the user has rated and characteristics of at least a portion of the rated items to which the user's ratings apply; 
 merge a plurality of the users' data that represents the patterns of dependency between ratings and characteristics for the plurality of users into a group preference profile wherein the merging comprises:
 identifying a plurality of consistent preference models for the users, 
 merging the identified consistent preference models into a merged preference model, and 
 including the merged preference model in the group preference profile; 
 
 receive a request for a group recommendation;
 select, from the database based on the group preference profile, a candidate item having characteristics that at least a plurality of users in group are expected to find appealing; and 
 
 generate a recommendation for the selected candidate item. 
 
 
     
     
       2. The system of  claim 1 , wherein the instructions that are configured instruct the processor to merge a plurality of the users' data into the group preference profile also comprise instructions to:
 identify merged item descriptions for a set of the rated items; and 
 include the merged item descriptions in the group preference profile. 
 
     
     
       3. The system of  claim 2 , wherein the instructions that are configured to instruct the processor to identify the merged item descriptions for the set of rated items comprise instructions to:
 identify similar item descriptors in the preference profiles for a plurality of the users; and 
 merge the similar item descriptors into the merged item description. 
 
     
     
       4. The system of  claim 3 , further comprising instructions that are configured to instruct the processor to require, as a condition of merging the similar item descriptors into the merged item description for an item, that at least a threshold portion of the users have preference profiles that include similar item descriptors for that item. 
     
     
       5. The system of  claim 1 , wherein the instructions that are configured instruct the processor to merge a plurality of the users' data into the group preference profile also comprise instructions to:
 identify merged degrees of appeal for a plurality of items and classes; and 
 include the merged degrees of appeal in the group preference profile. 
 
     
     
       6. The system of  claim 5 , wherein the instructions that are configured to instruct the processor to identify the merged degrees of appeal comprise instructions to:
 identify an item or class to which a plurality of the users have assigned similar ratings; and 
 merge the similar ratings into the merged degree of appeal. 
 
     
     
       7. The system of  claim 6 , further comprising instructions that are configured to instruct the processor to require, as a condition of merging the similar ratings into the merged degree of appeal, that at least a threshold portion of the users have assigned similar ratings to that item or class. 
     
     
       8. The system of  claim 1 , further comprising instructions that are configured to instruct the processor to determine a confidence level in the group preference profile. 
     
     
       9. The system of  claim 1 , further comprising instructions that are configured to instruct the processor to:
 determine whether the request for a recommendation comprises a constraint; and 
 if the request comprises a constraint, require that the candidate item satisfies the constraint before recommending the candidate item. 
 
     
     
       10. The system of  claim 1 , further comprising instructions that are configured to instruct the processor to:
 receive a precedence order for the users in the group; and 
 when developing the group preference profile, assign a higher weight to a pattern of dependency associated with a user who is higher in the precedence order than to a pattern of dependency associated with a user who is lower in the precedence order. 
 
     
     
       11. The system of  claim 1 , further comprising instructions that are configured to:
 include a plurality of additional merged preference models in the group preference profile; and 
 order the merged preference models that are included in the group preference profile on the basis of a social welfare function. 
 
     
     
       12. A method, comprising;
 receiving, via a user interface, a selection of a group having a plurality of members; 
 by a processor, identifying a preference profile for each member, wherein the preference profile for each member comprises:
 data that represents a pattern of dependency between the member's ratings for a plurality of items that the profile's member has rated and characteristics of at least a portion of the rated items to which the member's ratings apply; 
 
 by the processor, developing a group preference profile by merging the data that represents the patterns of dependency between ratings and characteristics for each of the members, wherein the merging comprises:
 identifying a plurality of consistent preference models for the members, wherein each of the consistent preference models is associated with a positive degree of appeal, 
 merging the identified consistent preference models into a merged preference model, and 
 including the merged preference model in the group preference profile; 
 
 receiving, via the user interface, a request for a recommendation for an item; 
 by the processor, accessing a database of candidate items, wherein each candidate item is associated with at least one characteristic; 
 by the processor, using the group preference profile to select, from the database, a candidate item having characteristics which are likely to appeal to at least a plurality of members of the group; and 
 by the processor, generating a recommendation for the selected candidate item. 
 
     
     
       13. The method of  claim 12 , further comprising:
 determining that the request for a recommendation comprises a constraint; and 
 when selecting the identified candidate item, confirming that the identified candidate item satisfies the constraint. 
 
     
     
       14. The method of  claim 12 , wherein merging the data that represents a pattern of dependency to develop the group preference profile also comprises:
 identifying merged item descriptions for a set of items that have been rated by the members; and 
 including the merged item descriptions in the group preference profile. 
 
     
     
       15. The method of  claim 14 , wherein identifying the merged item descriptions for the set of rated items comprises:
 identifying similar item descriptors in the preference profiles for a plurality of the users; and 
 merging the similar item descriptors into the merged item description. 
 
     
     
       16. The method of  claim 15 , further comprising, as a condition of merging the similar item descriptors into the merged item description for an item, requiring that at least a threshold portion of the members have preference profiles that include similar item descriptors for that item. 
     
     
       17. The method of  claim 12 , wherein merging the data that represents a pattern of dependency to develop the group preference profile also comprises:
 identifying merged degrees of appeal for a plurality of items and classes; and 
 including the merged degrees of appeal in the group preference profile. 
 
     
     
       18. The method of  claim 17 , wherein identifying the merged degrees of appeal comprises:
 identifying an item or class to which a plurality of the members have assigned similar ratings; and 
 merging the similar ratings into the merged degree of appeal. 
 
     
     
       19. The method of  claim 18 , further comprising, as a condition of merging the similar ratings into the merged degree of appeal, requiring that at least a threshold portion of the identified users have assigned similar ratings to that item or class. 
     
     
       20. The method of  claim 12 , further comprising determining a confidence level in the group preference profile. 
     
     
       21. The method of  claim 12 , further comprising:
 receiving a precedence order for the members; and 
 when developing the group preference profile, assigning a higher weight to a pattern of dependency associated with a member who is higher in the precedence order than to a pattern of dependency associated with a member who is lower in the precedence order. 
 
     
     
       22. The method of  claim 12 , further comprising:
 including a plurality of additional merged preference models in the group preference profile; and 
 ordering the merged preference models that are included in the group preference profile on the basis of a social welfare function. 
 
     
     
       23. A group recommendation system, comprising:
 a non-transitory computer-readable medium holding a database comprising characteristics for a plurality of candidate items; 
 a processor; and 
 a non-transitory computer-readable medium holding programming instructions that, when executed, are configured instruct the processor to:
 identify a preference profile for each user in a group of users, wherein the preference profile for each user comprises data that represents a pattern of dependency between the user's ratings for a plurality of items that the user has rated and characteristics of at least a portion of the rated items to which the user's ratings apply; 
 merge a plurality of the users' data that represents the patterns of dependency between ratings and characteristics for the plurality of users into a group preference profile wherein the merging comprises:
 identifying merged item descriptions for a set of the rated items by:
 identifying similar item descriptors in the preference profiles for a plurality of the users; and 
 merging the similar item descriptors into the merged item descriptions while requiring, as a condition of merging the similar item descriptors into the merged item description for an item, that at least a threshold portion of the users have preference profiles that include similar item descriptors for that item, and 
 
 including the merged item descriptions in the group preference profile; 
 
 receive a request for a group recommendation;
 select, from the database based on the group preference profile, a candidate item having characteristics that at least a plurality of users in group are expected to find appealing; and 
 
 generate a recommendation for the selected candidate item. 
 
 
     
     
       24. The system of  claim 23 , wherein the instructions that are configured to instruct the processor to merge a plurality of the users' data into the group preference profile also comprise instructions to:
 identify merged degrees of appeal for a plurality of items and classes; and 
 include the merged degrees of appeal in the group preference profile. 
 
     
     
       25. The system of  claim 24 , wherein the instructions that are configured to instruct the processor to identify the merged degrees of appeal comprise instructions to:
 identify an item or class to which a plurality of the users have assigned similar ratings; and 
 merge the similar ratings into the merged degree of appeal. 
 
     
     
       26. The system of  claim 25 , further comprising instructions that are configured to instruct the processor to require, as a condition of merging the similar ratings into the merged degree of appeal, that at least a threshold portion of the users have assigned similar ratings to that item or class. 
     
     
       27. The system of  claim 23 , further comprising instructions that are configured to instruct the processor to determine a confidence level in the group preference profile. 
     
     
       28. The system of  claim 23 , further comprising instructions that are configured to instruct the processor to:
 determine whether the request for a recommendation comprises a constraint; and 
 if the request comprises a constraint, require that the candidate item satisfies the constraint before recommending the candidate item. 
 
     
     
       29. The system of  claim 23 , further comprising instructions that are configured to instruct the processor to:
 receive a precedence order for the users in the group; and 
 when developing the group preference profile, assign a higher weight to a pattern of dependency associated with a user who is higher in the precedence order than to a pattern of dependency associated with a user who is lower in the precedence order. 
 
     
     
       30. The system of  claim 23 , wherein:
 the instructions that are configured instruct the processor to merge a plurality of the users' data into the group preference profile also comprise instructions to:
 identify a plurality of consistent preference models for the users, 
 merge the identified consistent preference models into a merged preference model, and 
 include the merged preference model in the group preference profile; and 
 
 the non-transitory storage medium further comprises additional instructions that are configured to:
 include a plurality of additional merged preference models in the group preference profile, and 
 order the merged preference models that are included in the group preference profile on the basis of a social welfare function. 
 
 
     
     
       31. A group recommendation system, comprising:
 a non-transitory computer-readable medium holding a database comprising characteristics for a plurality of candidate items; 
 a processor; and 
 a non-transitory computer-readable medium holding programming instructions that, when executed, are configured instruct the processor to:
 identify a preference profile for each user in a group of users, wherein the preference profile for each user comprises data that represents a pattern of dependency between the user's ratings for a plurality of items that the user has rated and characteristics of at least a portion of the rated items to which the user's ratings apply; 
 merge a plurality of the users' data that represents the patterns of dependency between ratings and characteristics for the plurality of users into a group preference profile wherein the merging comprises:
 identifying merged degrees of appeal for a plurality of items and classes by:
 identifying an item or class to which a plurality of the users have assigned similar ratings; and 
 merging the similar ratings into the merged degree of appeal while requiring, as a condition of merging the similar ratings into the merged degree of appeal, that at least a threshold portion of the users have assigned similar ratings to that item or class, and 
 
 including the merged degrees of appeal in the group preference profile; 
 
 receive a request for a group recommendation;
 select, from the database based on the group preference profile, a candidate item having characteristics that at least a plurality of users in group are expected to find appealing; and 
 
 generate a recommendation for the selected candidate item. 
 
 
     
     
       32. The system of  claim 31 , further comprising instructions that are configured to instruct the processor to determine a confidence level in the group preference profile. 
     
     
       33. The system of  claim 31 , further comprising instructions that are configured to instruct the processor to:
 determine whether the request for a recommendation comprises a constraint; and 
 if the request comprises a constraint, require that the candidate item satisfies the constraint before recommending the candidate item. 
 
     
     
       34. The system of  claim 31 , further comprising instructions that are configured to instruct the processor to:
 receive a precedence order for the users in the group; and 
 when developing the group preference profile, assign a higher weight to a pattern of dependency associated with a user who is higher in the precedence order than to a pattern of dependency associated with a user who is lower in the precedence order. 
 
     
     
       35. The system of  claim 31 , wherein:
 the instructions that are configured instruct the processor to merge a plurality of the users' data into the group preference profile also comprise instructions to:
 identify a plurality of consistent preference models for the users, 
 merge the identified consistent preference models into a merged preference model, and 
 include the merged preference model in the group preference profile; and 
 
 the non-transitory storage medium further comprises additional instructions that are configured to:
 include a plurality of additional merged preference models in the group preference profile, and 
 order the merged preference models that are included in the group preference profile on the basis of a social welfare function. 
 
 
     
     
       36. A method, comprising:
 receiving, via a user interface, a selection of a group having a plurality of members; 
 by a processor, identifying a preference profile for each member, wherein the preference profile for each member comprises data that represents a pattern of dependency between the member's ratings for a plurality of items that the profile's member has rated and characteristics of at least a portion of the rated items to which the member's ratings apply; 
 by the processor, developing a group preference profile by merging the data that represents the patterns of dependency between ratings and characteristics for each of the members, wherein the merging comprises:
 identifying merged item descriptions for a set of items that have been rated by the members by:
 identifying similar item descriptors in the preference profiles for a plurality of the members; and 
 merging the similar item descriptors into the merged item description while requiring, as a condition of merging the similar item descriptors into the merged item description for an item, that at least a threshold portion of the members have preference profiles that include similar item descriptors for that item, and 
 
 including the merged item descriptions in the group preference profile; 
 
 receiving, via the user interface, a request for a recommendation for an item; 
 by the processor, accessing a database of candidate items, wherein each candidate item is associated with at least one characteristic; 
 by the processor, using the group preference profile to select, from the database, a candidate item having characteristics which are likely to appeal to at least a plurality of members of the group; and 
 by the processor, generating a recommendation for the selected candidate item. 
 
     
     
       37. The method of  claim 36 , further comprising:
 determining that the request for a recommendation comprises a constraint; and 
 when selecting the identified candidate item, confirming that the identified candidate item satisfies the constraint. 
 
     
     
       38. The method of  claim 36 , wherein merging the data that represents a pattern of dependency to develop the group preference profile comprises:
 identifying merged degrees of appeal for a plurality of items and classes; and 
 including the merged degrees of appeal in the group preference profile. 
 
     
     
       39. The method of  claim 38 , wherein identifying each of the merged degrees of appeal comprises:
 identifying an item or class to which a plurality of the members have assigned similar ratings; and 
 merging the similar ratings into the merged degree of appeal. 
 
     
     
       40. The method of  claim 39 , further comprising, as a condition of merging the similar ratings into the merged degree of appeal, requiring that at least a threshold portion of the identified members have assigned similar ratings to that item or class. 
     
     
       41. The method of  claim 36 , further comprising determining a confidence level in the group preference profile. 
     
     
       42. The method of  claim 36 , further comprising:
 receiving a precedence order for the members; and 
 when developing the group preference profile, assigning a higher weight to a pattern of dependency associated with a member who is higher in the precedence order than to a pattern of dependency associated with a member who is lower in the precedence order. 
 
     
     
       43. The method of  claim 36 , wherein:
 merging the data that represents a pattern of dependency to develop the group preference profile also comprises:
 identifying a plurality of consistent preference models for the members, wherein each of the consistent preference models is associated with a positive degree of appeal, 
 merging the identified consistent preference models into a merged preference model, and 
 including the merged preference model in the group preference profile; and the method further comprises: 
 including a plurality of additional merged preference models in the group preference profile, and 
 ordering the merged preference models that are included in the group preference profile on the basis of a social welfare function. 
 
 
     
     
       44. A method, comprising:
 receiving, via a user interface, a selection of a group having a plurality of members; 
 by a processor, identifying a preference profile for each member, wherein the preference profile for each member comprises data that represents a pattern of dependency between the member's ratings for a plurality of items that the profile's member has rated and characteristics of at least a portion of the rated items to which the member's ratings apply; 
 by the processor, developing a group preference profile by merging the data that represents the patterns of dependency between ratings and characteristics for each of the members, wherein the merging comprises:
 identifying merged degrees of appeal for a plurality of items and classes by:
 identifying an item or class to which a plurality of the members have assigned similar ratings, and 
 merging the similar ratings into the merged degree of appeal, while requiring, as a condition of merging the similar ratings into the merged degree of appeal, requiring that at least a threshold portion of the identified members have assigned similar ratings to that item or class, and 
 
 including the merged degrees of appeal in the group preference profile; 
 
 receiving, via the user interface, a request for a recommendation for an item; 
 by the processor, accessing a database of candidate items, wherein each candidate item is associated with at least one characteristic; 
 by the processor, using the group preference profile to select, from a database, a candidate item having characteristics which are likely to appeal to at least a plurality of members of the group; and 
 by the processor, generating a recommendation for the selected candidate item. 
 
     
     
       45. The method of  claim 44 , further comprising determining a confidence level in the group preference profile. 
     
     
       46. The method of  claim 44 , further comprising:
 receiving a precedence order for the members; and 
 when developing the group preference profile, assigning a higher weight to a pattern of dependency associated with a member who is higher in the precedence order than to a pattern of dependency associated with a member who is lower in the precedence order. 
 
     
     
       47. The method of  claim 44 , wherein:
 merging the data that represents a pattern of dependency to develop the group preference profile also comprises:
 identifying a plurality of consistent preference models for the members, wherein each of the consistent preference models is associated with a positive degree of appeal, 
 merging the identified consistent preference models into a merged preference model, and 
 including the merged preference model in the group preference profile; and the method further comprises: 
 including a plurality of additional merged preference models in the group preference profile, and 
 ordering the merged preference models that are included in the group preference profile on the basis of a social welfare function.